src/third_party/libvpx/source/libvpx/vp9/common/mips/dspr2/vp9_itrans8_dspr2.c

   1 /*
   2  *  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
   3  *
   4  *  Use of this source code is governed by a BSD-style license
   5  *  that can be found in the LICENSE file in the root of the source
   6  *  tree. An additional intellectual property rights grant can be found
   7  *  in the file PATENTS.  All contributing project authors may
   8  *  be found in the AUTHORS file in the root of the source tree.
   9  */
  10
  11 #include <assert.h>
  12 #include <stdio.h>
  13
  14 #include "./vpx_config.h"
  15 #include "./vp9_rtcd.h"
  16 #include "vp9/common/vp9_common.h"
  17 #include "vp9/common/vp9_blockd.h"
  18 #include "vp9/common/vp9_idct.h"
  19 #include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
  20
  21 #if HAVE_DSPR2
  22 static void idct8_rows_dspr2(const int16_t *input, int16_t *output,
  23                              uint32_t no_rows) {
  24   int step1_0, step1_1, step1_2, step1_3, step1_4, step1_5, step1_6, step1_7;
  25   const int const_2_power_13 = 8192;
  26   int Temp0, Temp1, Temp2, Temp3, Temp4;
  27   int i;
  28
  29   for (i = no_rows; i--; ) {
  30     __asm__ __volatile__ (
  31         /*
  32           temp_1 = (input[0] + input[4]) * cospi_16_64;
  33           step2_0 = dct_const_round_shift(temp_1);
  34
  35           temp_2 = (input[0] - input[4]) * cospi_16_64;
  36           step2_1 = dct_const_round_shift(temp_2);
  37         */
  38         "lh       %[Temp0],             0(%[input])                     \n\t"
  39         "lh       %[Temp1],             8(%[input])                     \n\t"
  40         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
  41         "mthi     $zero,                $ac0                            \n\t"
  42         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
  43         "mthi     $zero,                $ac1                            \n\t"
  44         "add      %[Temp2],             %[Temp0],       %[Temp1]        \n\t"
  45         "madd     $ac0,                 %[Temp2],       %[cospi_16_64]  \n\t"
  46         "extp     %[Temp4],             $ac0,           31              \n\t"
  47
  48         "sub      %[Temp3],             %[Temp0],       %[Temp1]        \n\t"
  49         "madd     $ac1,                 %[Temp3],       %[cospi_16_64]  \n\t"
  50         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
  51         "mthi     $zero,                $ac0                            \n\t"
  52         "extp     %[Temp2],             $ac1,           31              \n\t"
  53
  54         /*
  55           temp_1 = input[2] * cospi_24_64 - input[6] * cospi_8_64;
  56           step2_2 = dct_const_round_shift(temp_1);
  57         */
  58         "lh       %[Temp0],             4(%[input])                     \n\t"
  59         "lh       %[Temp1],             12(%[input])                    \n\t"
  60         "madd     $ac0,                 %[Temp0],       %[cospi_24_64]  \n\t"
  61         "msub     $ac0,                 %[Temp1],       %[cospi_8_64]   \n\t"
  62         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
  63         "mthi     $zero,                $ac1                            \n\t"
  64         "extp     %[Temp3],             $ac0,           31              \n\t"
  65
  66         /*
  67           step1_1 = step2_1 + step2_2;
  68           step1_2 = step2_1 - step2_2;
  69         */
  70         "add      %[step1_1],           %[Temp2],       %[Temp3]        \n\t"
  71         "sub      %[step1_2],           %[Temp2],       %[Temp3]        \n\t"
  72
  73         /*
  74           temp_2 = input[2] * cospi_8_64 + input[6] * cospi_24_64;
  75           step2_3 = dct_const_round_shift(temp_2);
  76         */
  77         "madd     $ac1,                 %[Temp0],       %[cospi_8_64]   \n\t"
  78         "madd     $ac1,                 %[Temp1],       %[cospi_24_64]  \n\t"
  79         "extp     %[Temp1],             $ac1,           31              \n\t"
  80
  81         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
  82         "mthi     $zero,                $ac0                            \n\t"
  83
  84         /*
  85           step1_0 = step2_0 + step2_3;
  86           step1_3 = step2_0 - step2_3;
  87         */
  88         "add      %[step1_0],           %[Temp4],       %[Temp1]        \n\t"
  89         "sub      %[step1_3],           %[Temp4],       %[Temp1]        \n\t"
  90
  91         /*
  92           temp_1 = input[1] * cospi_28_64 - input[7] * cospi_4_64;
  93           step1_4 = dct_const_round_shift(temp_1);
  94         */
  95         "lh       %[Temp0],             2(%[input])                     \n\t"
  96         "madd     $ac0,                 %[Temp0],       %[cospi_28_64]  \n\t"
  97         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
  98         "mthi     $zero,                $ac1                            \n\t"
  99         "lh       %[Temp1],             14(%[input])                    \n\t"
 100         "lh       %[Temp0],             2(%[input])                     \n\t"
 101         "msub     $ac0,                 %[Temp1],       %[cospi_4_64]   \n\t"
 102         "extp     %[step1_4],           $ac0,           31              \n\t"
 103
 104         /*
 105           temp_2 = input[1] * cospi_4_64 + input[7] * cospi_28_64;
 106           step1_7 = dct_const_round_shift(temp_2);
 107         */
 108         "madd     $ac1,                 %[Temp0],       %[cospi_4_64]   \n\t"
 109         "madd     $ac1,                 %[Temp1],       %[cospi_28_64]  \n\t"
 110         "extp     %[step1_7],           $ac1,           31              \n\t"
 111
 112         /*
 113           temp_1 = input[5] * cospi_12_64 - input[3] * cospi_20_64;
 114           step1_5 = dct_const_round_shift(temp_1);
 115         */
 116         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 117         "mthi     $zero,                $ac0                            \n\t"
 118         "lh       %[Temp0],             10(%[input])                    \n\t"
 119         "madd     $ac0,                 %[Temp0],       %[cospi_12_64]  \n\t"
 120         "lh       %[Temp1],             6(%[input])                     \n\t"
 121         "msub     $ac0,                 %[Temp1],       %[cospi_20_64]  \n\t"
 122         "extp     %[step1_5],           $ac0,           31              \n\t"
 123
 124         /*
 125           temp_2 = input[5] * cospi_20_64 + input[3] * cospi_12_64;
 126           step1_6 = dct_const_round_shift(temp_2);
 127         */
 128         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 129         "mthi     $zero,                $ac1                            \n\t"
 130         "lh       %[Temp0],             10(%[input])                    \n\t"
 131         "madd     $ac1,                 %[Temp0],       %[cospi_20_64]  \n\t"
 132         "lh       %[Temp1],             6(%[input])                     \n\t"
 133         "madd     $ac1,                 %[Temp1],       %[cospi_12_64]  \n\t"
 134         "extp     %[step1_6],           $ac1,           31              \n\t"
 135
 136         /*
 137           temp_1 = (step1_7 - step1_6 - step1_4 + step1_5) * cospi_16_64;
 138           temp_2 = (step1_4 - step1_5 - step1_6 + step1_7) * cospi_16_64;
 139         */
 140         "sub      %[Temp0],             %[step1_7],     %[step1_6]      \n\t"
 141         "sub      %[Temp0],             %[Temp0],       %[step1_4]      \n\t"
 142         "add      %[Temp0],             %[Temp0],       %[step1_5]      \n\t"
 143         "sub      %[Temp1],             %[step1_4],     %[step1_5]      \n\t"
 144         "sub      %[Temp1],             %[Temp1],       %[step1_6]      \n\t"
 145         "add      %[Temp1],             %[Temp1],       %[step1_7]      \n\t"
 146
 147         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 148         "mthi     $zero,                $ac0                            \n\t"
 149         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 150         "mthi     $zero,                $ac1                            \n\t"
 151
 152         "madd     $ac0,                 %[Temp0],       %[cospi_16_64]  \n\t"
 153         "madd     $ac1,                 %[Temp1],       %[cospi_16_64]  \n\t"
 154
 155         /*
 156           step1_4 = step1_4 + step1_5;
 157           step1_7 = step1_6 + step1_7;
 158         */
 159         "add      %[step1_4],           %[step1_4],     %[step1_5]      \n\t"
 160         "add      %[step1_7],           %[step1_7],     %[step1_6]      \n\t"
 161
 162         "extp     %[step1_5],           $ac0,           31              \n\t"
 163         "extp     %[step1_6],           $ac1,           31              \n\t"
 164
 165         "add      %[Temp0],             %[step1_0],     %[step1_7]      \n\t"
 166         "sh       %[Temp0],             0(%[output])                    \n\t"
 167         "add      %[Temp1],             %[step1_1],     %[step1_6]      \n\t"
 168         "sh       %[Temp1],             16(%[output])                   \n\t"
 169         "add      %[Temp0],             %[step1_2],     %[step1_5]      \n\t"
 170         "sh       %[Temp0],             32(%[output])                   \n\t"
 171         "add      %[Temp1],             %[step1_3],     %[step1_4]      \n\t"
 172         "sh       %[Temp1],             48(%[output])                   \n\t"
 173
 174         "sub      %[Temp0],             %[step1_3],     %[step1_4]      \n\t"
 175         "sh       %[Temp0],             64(%[output])                   \n\t"
 176         "sub      %[Temp1],             %[step1_2],     %[step1_5]      \n\t"
 177         "sh       %[Temp1],             80(%[output])                   \n\t"
 178         "sub      %[Temp0],             %[step1_1],     %[step1_6]      \n\t"
 179         "sh       %[Temp0],             96(%[output])                   \n\t"
 180         "sub      %[Temp1],             %[step1_0],     %[step1_7]      \n\t"
 181         "sh       %[Temp1],             112(%[output])                  \n\t"
 182
 183         : [step1_0] "=&r" (step1_0), [step1_1] "=&r" (step1_1),
 184           [step1_2] "=&r" (step1_2), [step1_3] "=&r" (step1_3),
 185           [step1_4] "=&r" (step1_4), [step1_5] "=&r" (step1_5),
 186           [step1_6] "=&r" (step1_6), [step1_7] "=&r" (step1_7),
 187           [Temp0] "=&r" (Temp0), [Temp1] "=&r" (Temp1),
 188           [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3),
 189           [Temp4] "=&r" (Temp4)
 190         : [const_2_power_13] "r" (const_2_power_13),
 191           [cospi_16_64] "r" (cospi_16_64), [cospi_28_64] "r" (cospi_28_64),
 192           [cospi_4_64] "r" (cospi_4_64), [cospi_12_64] "r" (cospi_12_64),
 193           [cospi_20_64] "r" (cospi_20_64), [cospi_8_64] "r" (cospi_8_64),
 194           [cospi_24_64] "r" (cospi_24_64),
 195           [output] "r" (output), [input] "r" (input)
 196     );
 197
 198     input += 8;
 199     output += 1;
 200   }
 201 }
 202
 203 static void idct8_columns_add_blk_dspr2(int16_t *input, uint8_t *dest,
 204                                         int dest_stride) {
 205   int step1_0, step1_1, step1_2, step1_3, step1_4, step1_5, step1_6, step1_7;
 206   int Temp0, Temp1, Temp2, Temp3;
 207   int i;
 208   const int const_2_power_13 = 8192;
 209   uint8_t *dest_pix;
 210   uint8_t *cm = vp9_ff_cropTbl;
 211
 212   /* prefetch vp9_ff_cropTbl */
 213   vp9_prefetch_load(vp9_ff_cropTbl);
 214   vp9_prefetch_load(vp9_ff_cropTbl +  32);
 215   vp9_prefetch_load(vp9_ff_cropTbl +  64);
 216   vp9_prefetch_load(vp9_ff_cropTbl +  96);
 217   vp9_prefetch_load(vp9_ff_cropTbl + 128);
 218   vp9_prefetch_load(vp9_ff_cropTbl + 160);
 219   vp9_prefetch_load(vp9_ff_cropTbl + 192);
 220   vp9_prefetch_load(vp9_ff_cropTbl + 224);
 221
 222   for (i = 0; i < 8; ++i) {
 223       dest_pix = (dest + i);
 224
 225     __asm__ __volatile__ (
 226         /*
 227           temp_1 = (input[0] + input[4]) * cospi_16_64;
 228           step2_0 = dct_const_round_shift(temp_1);
 229
 230           temp_2 = (input[0] - input[4]) * cospi_16_64;
 231           step2_1 = dct_const_round_shift(temp_2);
 232         */
 233         "lh       %[Temp0],             0(%[input])                     \n\t"
 234         "lh       %[Temp1],             8(%[input])                     \n\t"
 235         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 236         "mthi     $zero,                $ac0                            \n\t"
 237         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 238         "mthi     $zero,                $ac1                            \n\t"
 239         "add      %[Temp2],             %[Temp0],       %[Temp1]        \n\t"
 240         "madd     $ac0,                 %[Temp2],       %[cospi_16_64]  \n\t"
 241         "extp     %[step1_6],           $ac0,           31              \n\t"
 242
 243         "sub      %[Temp3],             %[Temp0],       %[Temp1]        \n\t"
 244         "madd     $ac1,                 %[Temp3],       %[cospi_16_64]  \n\t"
 245         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 246         "mthi     $zero,                $ac0                            \n\t"
 247         "extp     %[Temp2],             $ac1,           31              \n\t"
 248
 249         /*
 250           temp_1 = input[2] * cospi_24_64 - input[6] * cospi_8_64;
 251           step2_2 = dct_const_round_shift(temp_1);
 252         */
 253         "lh       %[Temp0],             4(%[input])                     \n\t"
 254         "lh       %[Temp1],             12(%[input])                    \n\t"
 255         "madd     $ac0,                 %[Temp0],       %[cospi_24_64]  \n\t"
 256         "msub     $ac0,                 %[Temp1],       %[cospi_8_64]   \n\t"
 257         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 258         "mthi     $zero,                $ac1                            \n\t"
 259         "extp     %[Temp3],             $ac0,           31              \n\t"
 260
 261         /*
 262           step1_1 = step2_1 + step2_2;
 263           step1_2 = step2_1 - step2_2;
 264         */
 265         "add      %[step1_1],           %[Temp2],       %[Temp3]        \n\t"
 266         "sub      %[step1_2],           %[Temp2],       %[Temp3]        \n\t"
 267
 268         /*
 269           temp_2 = input[2] * cospi_8_64 + input[6] * cospi_24_64;
 270           step2_3 = dct_const_round_shift(temp_2);
 271         */
 272         "madd     $ac1,                 %[Temp0],       %[cospi_8_64]   \n\t"
 273         "madd     $ac1,                 %[Temp1],       %[cospi_24_64]  \n\t"
 274         "extp     %[Temp1],             $ac1,           31              \n\t"
 275
 276         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 277         "mthi     $zero,                $ac0                            \n\t"
 278
 279         /*
 280           step1_0 = step2_0 + step2_3;
 281           step1_3 = step2_0 - step2_3;
 282         */
 283         "add      %[step1_0],           %[step1_6],     %[Temp1]        \n\t"
 284         "sub      %[step1_3],           %[step1_6],     %[Temp1]        \n\t"
 285
 286         /*
 287           temp_1 = input[1] * cospi_28_64 - input[7] * cospi_4_64;
 288           step1_4 = dct_const_round_shift(temp_1);
 289         */
 290         "lh       %[Temp0],             2(%[input])                     \n\t"
 291         "madd     $ac0,                 %[Temp0],       %[cospi_28_64]  \n\t"
 292         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 293         "mthi     $zero,                $ac1                            \n\t"
 294         "lh       %[Temp1],             14(%[input])                    \n\t"
 295         "lh       %[Temp0],             2(%[input])                     \n\t"
 296         "msub     $ac0,                 %[Temp1],       %[cospi_4_64]   \n\t"
 297         "extp     %[step1_4],           $ac0,           31              \n\t"
 298
 299         /*
 300           temp_2 = input[1] * cospi_4_64 + input[7] * cospi_28_64;
 301           step1_7 = dct_const_round_shift(temp_2);
 302         */
 303         "madd     $ac1,                 %[Temp0],       %[cospi_4_64]   \n\t"
 304         "madd     $ac1,                 %[Temp1],       %[cospi_28_64]  \n\t"
 305         "extp     %[step1_7],           $ac1,           31              \n\t"
 306
 307         /*
 308           temp_1 = input[5] * cospi_12_64 - input[3] * cospi_20_64;
 309           step1_5 = dct_const_round_shift(temp_1);
 310         */
 311         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 312         "mthi     $zero,                $ac0                            \n\t"
 313         "lh       %[Temp0],             10(%[input])                    \n\t"
 314         "madd     $ac0,                 %[Temp0],       %[cospi_12_64]  \n\t"
 315         "lh       %[Temp1],             6(%[input])                     \n\t"
 316         "msub     $ac0,                 %[Temp1],       %[cospi_20_64]  \n\t"
 317         "extp     %[step1_5],           $ac0,           31              \n\t"
 318
 319         /*
 320           temp_2 = input[5] * cospi_20_64 + input[3] * cospi_12_64;
 321           step1_6 = dct_const_round_shift(temp_2);
 322         */
 323         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 324         "mthi     $zero,                $ac1                            \n\t"
 325         "lh       %[Temp0],             10(%[input])                    \n\t"
 326         "madd     $ac1,                 %[Temp0],       %[cospi_20_64]  \n\t"
 327         "lh       %[Temp1],             6(%[input])                     \n\t"
 328         "madd     $ac1,                 %[Temp1],       %[cospi_12_64]  \n\t"
 329         "extp     %[step1_6],           $ac1,           31              \n\t"
 330
 331         /*
 332           temp_1 = (step1_7 - step1_6 - step1_4 + step1_5) * cospi_16_64;
 333           temp_2 = (step1_4 - step1_5 - step1_6 + step1_7) * cospi_16_64;
 334         */
 335         "sub      %[Temp0],             %[step1_7],     %[step1_6]      \n\t"
 336         "sub      %[Temp0],             %[Temp0],       %[step1_4]      \n\t"
 337         "add      %[Temp0],             %[Temp0],       %[step1_5]      \n\t"
 338         "sub      %[Temp1],             %[step1_4],     %[step1_5]      \n\t"
 339         "sub      %[Temp1],             %[Temp1],       %[step1_6]      \n\t"
 340         "add      %[Temp1],             %[Temp1],       %[step1_7]      \n\t"
 341
 342         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 343         "mthi     $zero,                $ac0                            \n\t"
 344         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 345         "mthi     $zero,                $ac1                            \n\t"
 346
 347         "madd     $ac0,                 %[Temp0],       %[cospi_16_64]  \n\t"
 348         "madd     $ac1,                 %[Temp1],       %[cospi_16_64]  \n\t"
 349
 350         /*
 351           step1_4 = step1_4 + step1_5;
 352           step1_7 = step1_6 + step1_7;
 353         */
 354         "add      %[step1_4],           %[step1_4],     %[step1_5]      \n\t"
 355         "add      %[step1_7],           %[step1_7],     %[step1_6]      \n\t"
 356
 357         "extp     %[step1_5],           $ac0,           31              \n\t"
 358         "extp     %[step1_6],           $ac1,           31              \n\t"
 359
 360         /* add block */
 361         "lbu      %[Temp1],             0(%[dest_pix])                  \n\t"
 362         "add      %[Temp0],             %[step1_0],     %[step1_7]      \n\t"
 363         "addi     %[Temp0],             %[Temp0],       16              \n\t"
 364         "sra      %[Temp0],             %[Temp0],       5               \n\t"
 365         "add      %[Temp1],             %[Temp1],       %[Temp0]        \n\t"
 366         "add      %[Temp0],             %[step1_1],     %[step1_6]      \n\t"
 367         "lbux     %[Temp2],             %[Temp1](%[cm])                 \n\t"
 368         "sb       %[Temp2],             0(%[dest_pix])                  \n\t"
 369         "addu     %[dest_pix],          %[dest_pix],    %[dest_stride]  \n\t"
 370
 371         "lbu      %[Temp1],             0(%[dest_pix])                  \n\t"
 372         "addi     %[Temp0],             %[Temp0],       16              \n\t"
 373         "sra      %[Temp0],             %[Temp0],       5               \n\t"
 374         "add      %[Temp1],             %[Temp1],       %[Temp0]        \n\t"
 375         "add      %[Temp0],             %[step1_2],     %[step1_5]      \n\t"
 376         "lbux     %[Temp2],             %[Temp1](%[cm])                 \n\t"
 377         "sb       %[Temp2],             0(%[dest_pix])                  \n\t"
 378         "addu     %[dest_pix],          %[dest_pix],    %[dest_stride]  \n\t"
 379
 380         "lbu      %[Temp1],             0(%[dest_pix])                  \n\t"
 381         "addi     %[Temp0],             %[Temp0],       16              \n\t"
 382         "sra      %[Temp0],             %[Temp0],       5               \n\t"
 383         "add      %[Temp1],             %[Temp1],       %[Temp0]        \n\t"
 384         "add      %[Temp0],             %[step1_3],     %[step1_4]      \n\t"
 385         "lbux     %[Temp2],             %[Temp1](%[cm])                 \n\t"
 386         "sb       %[Temp2],             0(%[dest_pix])                  \n\t"
 387         "addu     %[dest_pix],          %[dest_pix],    %[dest_stride]  \n\t"
 388
 389         "lbu      %[Temp1],             0(%[dest_pix])                  \n\t"
 390         "addi     %[Temp0],             %[Temp0],       16              \n\t"
 391         "sra      %[Temp0],             %[Temp0],       5               \n\t"
 392         "add      %[Temp1],             %[Temp1],       %[Temp0]        \n\t"
 393         "sub      %[Temp0],             %[step1_3],     %[step1_4]      \n\t"
 394         "lbux     %[Temp2],             %[Temp1](%[cm])                 \n\t"
 395         "sb       %[Temp2],             0(%[dest_pix])                  \n\t"
 396         "addu     %[dest_pix],          %[dest_pix],    %[dest_stride]  \n\t"
 397
 398         "lbu      %[Temp1],             0(%[dest_pix])                  \n\t"
 399         "addi     %[Temp0],             %[Temp0],       16              \n\t"
 400         "sra      %[Temp0],             %[Temp0],       5               \n\t"
 401         "add      %[Temp1],             %[Temp1],       %[Temp0]        \n\t"
 402         "sub      %[Temp0],             %[step1_2],     %[step1_5]      \n\t"
 403         "lbux     %[Temp2],             %[Temp1](%[cm])                 \n\t"
 404         "sb       %[Temp2],             0(%[dest_pix])                  \n\t"
 405         "addu     %[dest_pix],          %[dest_pix],    %[dest_stride]  \n\t"
 406
 407         "lbu      %[Temp1],             0(%[dest_pix])                  \n\t"
 408         "addi     %[Temp0],             %[Temp0],       16              \n\t"
 409         "sra      %[Temp0],             %[Temp0],       5               \n\t"
 410         "add      %[Temp1],             %[Temp1],       %[Temp0]        \n\t"
 411         "sub      %[Temp0],             %[step1_1],     %[step1_6]      \n\t"
 412         "lbux     %[Temp2],             %[Temp1](%[cm])                 \n\t"
 413         "sb       %[Temp2],             0(%[dest_pix])                  \n\t"
 414         "addu     %[dest_pix],          %[dest_pix],    %[dest_stride]  \n\t"
 415
 416         "lbu      %[Temp1],             0(%[dest_pix])                  \n\t"
 417         "addi     %[Temp0],             %[Temp0],       16              \n\t"
 418         "sra      %[Temp0],             %[Temp0],       5               \n\t"
 419         "add      %[Temp1],             %[Temp1],       %[Temp0]        \n\t"
 420         "sub      %[Temp0],             %[step1_0],     %[step1_7]      \n\t"
 421         "lbux     %[Temp2],             %[Temp1](%[cm])                 \n\t"
 422         "sb       %[Temp2],             0(%[dest_pix])                  \n\t"
 423         "addu     %[dest_pix],          %[dest_pix],    %[dest_stride]  \n\t"
 424
 425         "lbu      %[Temp1],             0(%[dest_pix])                  \n\t"
 426         "addi     %[Temp0],             %[Temp0],       16              \n\t"
 427         "sra      %[Temp0],             %[Temp0],       5               \n\t"
 428         "add      %[Temp1],             %[Temp1],       %[Temp0]        \n\t"
 429         "lbux     %[Temp2],             %[Temp1](%[cm])                 \n\t"
 430         "sb       %[Temp2],             0(%[dest_pix])                  \n\t"
 431
 432         : [step1_0] "=&r" (step1_0), [step1_1] "=&r" (step1_1),
 433           [step1_2] "=&r" (step1_2), [step1_3] "=&r" (step1_3),
 434           [step1_4] "=&r" (step1_4), [step1_5] "=&r" (step1_5),
 435           [step1_6] "=&r" (step1_6), [step1_7] "=&r" (step1_7),
 436           [Temp0] "=&r" (Temp0), [Temp1] "=&r" (Temp1),
 437           [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3),
 438           [dest_pix] "+r" (dest_pix)
 439         : [const_2_power_13] "r" (const_2_power_13),
 440           [cospi_16_64] "r" (cospi_16_64), [cospi_28_64] "r" (cospi_28_64),
 441           [cospi_4_64] "r" (cospi_4_64), [cospi_12_64] "r" (cospi_12_64),
 442           [cospi_20_64] "r" (cospi_20_64), [cospi_8_64] "r" (cospi_8_64),
 443           [cospi_24_64] "r" (cospi_24_64),
 444           [input] "r" (input), [cm] "r" (cm), [dest_stride] "r" (dest_stride)
 445     );
 446
 447     input += 8;
 448   }
 449 }
 450
 451 void vp9_idct8x8_64_add_dspr2(const int16_t *input, uint8_t *dest,
 452                               int dest_stride) {
 453   DECLARE_ALIGNED(32, int16_t, out[8 * 8]);
 454   int16_t *outptr = out;
 455   uint32_t pos = 45;
 456
 457   /* bit positon for extract from acc */
 458   __asm__ __volatile__ (
 459     "wrdsp    %[pos],    1    \n\t"
 460     :
 461     : [pos] "r" (pos)
 462   );
 463
 464   // First transform rows
 465   idct8_rows_dspr2(input, outptr, 8);
 466
 467   // Then transform columns and add to dest
 468   idct8_columns_add_blk_dspr2(&out[0], dest, dest_stride);
 469 }
 470
 471 static void iadst8_dspr2(const int16_t *input, int16_t *output) {
 472   int s0, s1, s2, s3, s4, s5, s6, s7;
 473   int x0, x1, x2, x3, x4, x5, x6, x7;
 474
 475   x0 = input[7];
 476   x1 = input[0];
 477   x2 = input[5];
 478   x3 = input[2];
 479   x4 = input[3];
 480   x5 = input[4];
 481   x6 = input[1];
 482   x7 = input[6];
 483
 484   if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7)) {
 485     output[0] = output[1] = output[2] = output[3] = output[4]
 486               = output[5] = output[6] = output[7] = 0;
 487     return;
 488   }
 489
 490   // stage 1
 491   s0 = cospi_2_64  * x0 + cospi_30_64 * x1;
 492   s1 = cospi_30_64 * x0 - cospi_2_64  * x1;
 493   s2 = cospi_10_64 * x2 + cospi_22_64 * x3;
 494   s3 = cospi_22_64 * x2 - cospi_10_64 * x3;
 495   s4 = cospi_18_64 * x4 + cospi_14_64 * x5;
 496   s5 = cospi_14_64 * x4 - cospi_18_64 * x5;
 497   s6 = cospi_26_64 * x6 + cospi_6_64  * x7;
 498   s7 = cospi_6_64  * x6 - cospi_26_64 * x7;
 499
 500   x0 = ROUND_POWER_OF_TWO((s0 + s4), DCT_CONST_BITS);
 501   x1 = ROUND_POWER_OF_TWO((s1 + s5), DCT_CONST_BITS);
 502   x2 = ROUND_POWER_OF_TWO((s2 + s6), DCT_CONST_BITS);
 503   x3 = ROUND_POWER_OF_TWO((s3 + s7), DCT_CONST_BITS);
 504   x4 = ROUND_POWER_OF_TWO((s0 - s4), DCT_CONST_BITS);
 505   x5 = ROUND_POWER_OF_TWO((s1 - s5), DCT_CONST_BITS);
 506   x6 = ROUND_POWER_OF_TWO((s2 - s6), DCT_CONST_BITS);
 507   x7 = ROUND_POWER_OF_TWO((s3 - s7), DCT_CONST_BITS);
 508
 509   // stage 2
 510   s0 = x0;
 511   s1 = x1;
 512   s2 = x2;
 513   s3 = x3;
 514   s4 =  cospi_8_64  * x4 + cospi_24_64 * x5;
 515   s5 =  cospi_24_64 * x4 - cospi_8_64  * x5;
 516   s6 = -cospi_24_64 * x6 + cospi_8_64  * x7;
 517   s7 =  cospi_8_64  * x6 + cospi_24_64 * x7;
 518
 519   x0 = s0 + s2;
 520   x1 = s1 + s3;
 521   x2 = s0 - s2;
 522   x3 = s1 - s3;
 523   x4 = ROUND_POWER_OF_TWO((s4 + s6), DCT_CONST_BITS);
 524   x5 = ROUND_POWER_OF_TWO((s5 + s7), DCT_CONST_BITS);
 525   x6 = ROUND_POWER_OF_TWO((s4 - s6), DCT_CONST_BITS);
 526   x7 = ROUND_POWER_OF_TWO((s5 - s7), DCT_CONST_BITS);
 527
 528   // stage 3
 529   s2 = cospi_16_64 * (x2 + x3);
 530   s3 = cospi_16_64 * (x2 - x3);
 531   s6 = cospi_16_64 * (x6 + x7);
 532   s7 = cospi_16_64 * (x6 - x7);
 533
 534   x2 = ROUND_POWER_OF_TWO((s2), DCT_CONST_BITS);
 535   x3 = ROUND_POWER_OF_TWO((s3), DCT_CONST_BITS);
 536   x6 = ROUND_POWER_OF_TWO((s6), DCT_CONST_BITS);
 537   x7 = ROUND_POWER_OF_TWO((s7), DCT_CONST_BITS);
 538
 539   output[0] =  x0;
 540   output[1] = -x4;
 541   output[2] =  x6;
 542   output[3] = -x2;
 543   output[4] =  x3;
 544   output[5] = -x7;
 545   output[6] =  x5;
 546   output[7] = -x1;
 547 }
 548
 549 void vp9_iht8x8_64_add_dspr2(const int16_t *input, uint8_t *dest,
 550                              int dest_stride, int tx_type) {
 551   int i, j;
 552   DECLARE_ALIGNED(32, int16_t, out[8 * 8]);
 553   int16_t *outptr = out;
 554   int16_t temp_in[8 * 8], temp_out[8];
 555   uint32_t pos = 45;
 556
 557   /* bit positon for extract from acc */
 558   __asm__ __volatile__ (
 559     "wrdsp    %[pos],    1    \n\t"
 560     :
 561     : [pos] "r" (pos)
 562   );
 563
 564   switch (tx_type) {
 565     case DCT_DCT:     // DCT in both horizontal and vertical
 566       idct8_rows_dspr2(input, outptr, 8);
 567       idct8_columns_add_blk_dspr2(&out[0], dest, dest_stride);
 568       break;
 569     case ADST_DCT:    // ADST in vertical, DCT in horizontal
 570       idct8_rows_dspr2(input, outptr, 8);
 571
 572       for (i = 0; i < 8; ++i) {
 573         iadst8_dspr2(&out[i * 8], temp_out);
 574
 575         for (j = 0; j < 8; ++j)
 576           dest[j * dest_stride + i] =
 577                     clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
 578                                       + dest[j * dest_stride + i]);
 579       }
 580       break;
 581     case DCT_ADST:    // DCT in vertical, ADST in horizontal
 582       for (i = 0; i < 8; ++i) {
 583         iadst8_dspr2(input, outptr);
 584         input += 8;
 585         outptr += 8;
 586       }
 587
 588       for (i = 0; i < 8; ++i) {
 589         for (j = 0; j < 8; ++j) {
 590           temp_in[i * 8 + j] = out[j * 8 + i];
 591         }
 592       }
 593       idct8_columns_add_blk_dspr2(&temp_in[0], dest, dest_stride);
 594       break;
 595     case ADST_ADST:   // ADST in both directions
 596       for (i = 0; i < 8; ++i) {
 597         iadst8_dspr2(input, outptr);
 598         input += 8;
 599         outptr += 8;
 600       }
 601
 602       for (i = 0; i < 8; ++i) {
 603         for (j = 0; j < 8; ++j)
 604           temp_in[j] = out[j * 8 + i];
 605
 606         iadst8_dspr2(temp_in, temp_out);
 607
 608         for (j = 0; j < 8; ++j)
 609           dest[j * dest_stride + i] =
 610                 clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
 611                                       + dest[j * dest_stride + i]);
 612       }
 613       break;
 614     default:
 615       printf("vp9_short_iht8x8_add_dspr2 : Invalid tx_type\n");
 616       break;
 617   }
 618 }
 619
 620 void vp9_idct8x8_10_add_dspr2(const int16_t *input, uint8_t *dest,
 621                               int dest_stride) {
 622   DECLARE_ALIGNED(32, int16_t, out[8 * 8]);
 623   int16_t *outptr = out;
 624   uint32_t pos = 45;
 625
 626   /* bit positon for extract from acc */
 627   __asm__ __volatile__ (
 628     "wrdsp    %[pos],    1    \n\t"
 629     :
 630     : [pos] "r" (pos)
 631   );
 632
 633   // First transform rows
 634   idct8_rows_dspr2(input, outptr, 4);
 635
 636   outptr += 4;
 637
 638   __asm__ __volatile__ (
 639       "sw  $zero,   0(%[outptr])  \n\t"
 640       "sw  $zero,   4(%[outptr])  \n\t"
 641       "sw  $zero,  16(%[outptr])  \n\t"
 642       "sw  $zero,  20(%[outptr])  \n\t"
 643       "sw  $zero,  32(%[outptr])  \n\t"
 644       "sw  $zero,  36(%[outptr])  \n\t"
 645       "sw  $zero,  48(%[outptr])  \n\t"
 646       "sw  $zero,  52(%[outptr])  \n\t"
 647       "sw  $zero,  64(%[outptr])  \n\t"
 648       "sw  $zero,  68(%[outptr])  \n\t"
 649       "sw  $zero,  80(%[outptr])  \n\t"
 650       "sw  $zero,  84(%[outptr])  \n\t"
 651       "sw  $zero,  96(%[outptr])  \n\t"
 652       "sw  $zero, 100(%[outptr])  \n\t"
 653       "sw  $zero, 112(%[outptr])  \n\t"
 654       "sw  $zero, 116(%[outptr])  \n\t"
 655
 656       :
 657       : [outptr] "r" (outptr)
 658   );
 659
 660
 661   // Then transform columns and add to dest
 662   idct8_columns_add_blk_dspr2(&out[0], dest, dest_stride);
 663 }
 664
 665 void vp9_idct8x8_1_add_dspr2(const int16_t *input, uint8_t *dest,
 666                              int dest_stride) {
 667   uint32_t pos = 45;
 668   int32_t out;
 669   int32_t r;
 670   int32_t a1, absa1;
 671   int32_t t1, t2, vector_a1, vector_1, vector_2;
 672
 673   /* bit positon for extract from acc */
 674   __asm__ __volatile__ (
 675     "wrdsp      %[pos],     1           \n\t"
 676
 677     :
 678     : [pos] "r" (pos)
 679   );
 680
 681   out = DCT_CONST_ROUND_SHIFT_TWICE_COSPI_16_64(input[0]);
 682   __asm__ __volatile__ (
 683       "addi     %[out],     %[out],     16      \n\t"
 684       "sra      %[a1],      %[out],     5       \n\t"
 685
 686       : [out] "+r" (out), [a1] "=r" (a1)
 687       :
 688   );
 689
 690   if (a1 < 0) {
 691     /* use quad-byte
 692      * input and output memory are four byte aligned */
 693     __asm__ __volatile__ (
 694         "abs        %[absa1],       %[a1]       \n\t"
 695         "replv.qb   %[vector_a1],   %[absa1]    \n\t"
 696
 697         : [absa1] "=r" (absa1), [vector_a1] "=r" (vector_a1)
 698         : [a1] "r" (a1)
 699     );
 700
 701     for (r = 8; r--;) {
 702       __asm__ __volatile__ (
 703           "lw           %[t1],          0(%[dest])                      \n\t"
 704           "lw           %[t2],          4(%[dest])                      \n\t"
 705           "subu_s.qb    %[vector_1],    %[t1],          %[vector_a1]    \n\t"
 706           "subu_s.qb    %[vector_2],    %[t2],          %[vector_a1]    \n\t"
 707           "sw           %[vector_1],    0(%[dest])                      \n\t"
 708           "sw           %[vector_2],    4(%[dest])                      \n\t"
 709           "add          %[dest],        %[dest],        %[dest_stride]  \n\t"
 710
 711           : [t1] "=&r" (t1), [t2] "=&r" (t2),
 712             [vector_1] "=&r" (vector_1), [vector_2] "=&r" (vector_2),
 713             [dest] "+&r" (dest)
 714           : [dest_stride] "r" (dest_stride), [vector_a1] "r" (vector_a1)
 715       );
 716     }
 717   } else {
 718     /* use quad-byte
 719      * input and output memory are four byte aligned */
 720     __asm__ __volatile__ (
 721         "replv.qb   %[vector_a1],   %[a1]   \n\t"
 722
 723         : [vector_a1] "=r" (vector_a1)
 724         : [a1] "r" (a1)
 725     );
 726
 727     for (r = 8; r--;) {
 728       __asm__ __volatile__ (
 729           "lw           %[t1],          0(%[dest])                      \n\t"
 730           "lw           %[t2],          4(%[dest])                      \n\t"
 731           "addu_s.qb    %[vector_1],    %[t1],          %[vector_a1]    \n\t"
 732           "addu_s.qb    %[vector_2],    %[t2],          %[vector_a1]    \n\t"
 733           "sw           %[vector_1],    0(%[dest])                      \n\t"
 734           "sw           %[vector_2],    4(%[dest])                      \n\t"
 735           "add          %[dest],        %[dest],        %[dest_stride]  \n\t"
 736
 737           : [t1] "=&r" (t1), [t2] "=&r" (t2),
 738             [vector_1] "=&r" (vector_1), [vector_2] "=&r" (vector_2),
 739             [dest] "+r" (dest)
 740           : [dest_stride] "r" (dest_stride), [vector_a1] "r" (vector_a1)
 741       );
 742     }
 743   }
 744 }
 745 #endif  // #if HAVE_DSPR2